SmartDelivery · Machine Learning and the Internet of Things for Optimisation of the Last Mile Delivery

Machine Learning and the Internet of Things for Optimisation of the Last Mile Delivery

Last-mile delivery (LMD) is the final stage of the logistics process, where goods move from a hub or distribution center to the customer. It is widely recognized as the most expensive and operationally challenging segment of the logistics chain, accounting for up to 41% of total logistics costs. Beyond direct operational expenses (fuel, maintenance, wages), LMD generates significant negative externalities, including urban congestion, accidents, pollution, and infrastructure wear. The rapid expansion of e-commerce, accelerated by post-pandemic shifts in consumer behavior, has intensified the urgency of addressing the inefficiencies of the LMD process. This aligns with the European Green Deal and UN Sustainable Development Goals (SDGs 3 and 11), which aim to foster sustainable mobility and improve citizens' quality of life. From an operational research perspective, route planning in LMD is typically modeled, in its baseline academic form, through the Capacitated Vehicle Routing Problem (CVRP), and scientific literature has proposed several novel methodologies to solve CVRP instances. However, these theoretical advances often fail to produce the expected results in real-world scenarios. This is largely caused by inaccurate travel data estimates (e.g., derived from online open maps) that fail to account for the unpredictable events occurring during deliveries. Furthermore, the selection of the most suitable solver for specific problem instances is frequently manual and biased. Moreover, standard solvers typically ignore the "human factor" (specifically, driver familiarity with delivery zones) which directly affects service times and customer experience. The SmartDelivery project aims to bridge these gaps by developing a novel, layered platform for LMD optimization structured around three core technical objectives: (1) the design of a Machine Learning-based algorithm selection module to automatically choose the most suitable heuristic or metaheuristic solver from a portfolio based on instance features and performance metrics, overcoming the limitations of manual selection; (2) the implementation of a robust methodology to replace inaccurate, static travel time data with real-time estimations via custom IoT devices, ensuring high-fidelity inputs for the routing process; and (3) the development of a novel "sixth sense" concept to dynamically assign routes to drivers based on their efficiency and familiarity with delivery zones, replacing standard random assignment strategies and significantly reducing service times. By integrating Machine Learning (ML) techniques with real-world sensing, SmartDelivery aims to significantly enhance LMD efficiency, thereby reducing operational costs for logistics companies and mitigating environmental externalities, thus contributing to the realization of smarter, greener, and more efficient European cities.

Текст от CORDIS, на английски · Данни: CORDIS, © Европейски съюз